Tracking for inspection in energy transmission power lines using unmanned aerial vehicles: a systematic review of current and specific literature

Received Mar 22, 2020 Revised Apr 29, 2020 Accepted Aug 18, 2020 Power transmission lines are of great importance for the operation of all sectors of society, such as commerce, industry and public agencies. To ensure reliability and availability of power supply, regular and occasional inspections are conducted, mostly using patrol with binoculars, helicopters, or truck cranes. Research is being developed using unmanned aerial vehicle (UAV) to make this activity autonomous, faster, safer, and less costly. The present work aims to analyze research related to the autonomous control of the UAV along the transmission lines through a systematic review of the literature (SRL), apply a viable solution and to verify the possible lacuna in this state of the art. Improvements in safety, computational process and energy efficiency with low-cost were identified. The results presented can help the research community to perform the working this state of art, from the suggestions of autonomous tracking of transmission lines.

The aim of this work is to identify in state of the art, in publications the best way to perform power transmission line tracking using UAV. For this, the systematic literature review (SRL) process was used, verifying in the academy how this problem is being treated and providing a technical analysis for the solution. The SRL applies a well-designed research protocol, to search for articles in scientific databases, and with result of the analysis, answer the research questions. This study aims to assist other researchers and professionals in verifying the methods used, supporting decision making in future projects. There is great importance in the periodic review of these facilities, both for society, and for the possibility of automating the process offering low cost and high efficiency, thus providing the stability of the energy service.
To assist in the development of this study, the Parsifal [3] tool was used. It has the main steps necessary for the development of SRL, in addition to enabling online collaboration of the authors and the generation of analytical reports. Thus, an extensive material is produced, with all the necessary information of the articles, to synthesize the result in this work.
This paper is structured as follows. Section 2 describes the research methodology used in this work, the processes performed, and the research questions to be answered from this systematic review. Section 3 presents the analysis of the selected articles and their classifications. Section 4 presents the results obtained in the extraction. Section 5 there was a discussion of the results regarding the research questions and the last section presents the conclusion of this work.

RESEARCH METHOD
The methodology chosen for this study follows the detailed protocol originally proposed by Kitchenham [4] presented in Figure 1. This model has a methodologically well-defined research process to be applied that identifies, analyzes, and interprets available evidence related to research questions.

Research definition
The objective of this work is to analyze UAV guidance techniques for the inspection of power transmission lines and to identify materials and methods used in the autonomous transmission lines inspection from the point of view of the researchers in the area in the academic context. The elaboration of the question is of paramount importance for the outcome of the SRL. The answers to these questions make possible the specific analysis to determine the initial focus of the research and make possible the contribution to the academic environment. Thus, the following main question (MQ) is defined: -What are the best techniques used to trace transmission lines in an automated, safe and low-cost way? The MQ seeks to obtain the best way to perform the transmission line tracking and, for this, it takes into consideration important aspects such as efficiency, safety and cost.
From the MQ, specific questions (SQ) are defined in a secondary way: -SQ1: What is the cost of implementing the materials and techniques used to inspect the transmission lines? -SQ2: What is the computational and energetic efficiency of the techniques used to guide the UAV? -SQ3: What is the level of security and the locations of the implemented solutions for the tracking of transmission lines? Subsequently, the evidence will be materialized in published primary articles retrieved from selected research sources. The retrieval of these works happens through a search string (SS) that summarizes the questions to be searched. Thus, criteria were defined to guarantee the balance, feasibility and accessibility of the information for the execution of the study through this string, the criteria for conduction are: a. Possessing search engine that allows the use of logical expressions or equivalent mechanism, returning the expected from the string. b. Include in its collection with relevance in the academic environment in the areas of computing and engineering. c. The search engines should allow search in the full text of the publications. d. Own publication date between January 1, 2014 and June 1, 2019. e. It should have the feature of filtering only article. f. The publication must be available in the English language.
The research was conducted from four digital libraries; they are Association for Computing Machinery (ACM) (dl.acm.org), Institute of Electrical and Electronic Engineers (IEEE) (www.ieee.org), Scopus (www.scopus.com) and Compendex (www.engineeringvillage.com). These bases were chosen by the experiment carried out by Dybala et al. [5]. As argued above, some databases currently known as Google Scholar, PubMed, and dissertation and theses libraries have not been considered, as they do not meet the driving criteria cited. However, it is known that most of the articles contained in these platforms are also available in the selected bases, as these are from the field of study and are of high academic relevance.

Search terms
In the context of the study objective, this work defines a strategy to construct the SS, this string is represented generally by the keywords related to the search questions and that can be used in the search. The strategy included: a. Identification of keywords considering the main research question. b. Identification of keywords considering the control article [6] originated from the research group. c. Synonyms identified in the context. The main search expression needs specific adjustments in the string to meet the driving criteria. Following the reapplication of the strategy of construction and refinement of the string the final expression used was: ("Autonomous" OR "Self-Navigation" OR "Self Navigation" OR Automatic OR AutoPilot) AND ("Remotely Piloted Aircraft" OR RPAS OR ARP OR "Unmanned Aerial Vehicle" OR UAV OR UAS OR DRONE OR "unmanned aerial system" OR Airplane) AND (Imaging OR Inspection OR Maintenance) AND ("Transmission Power" OR "Transmission Line" OR "Power line transmission" OR "Power Network" OR "Power line inspection" OR "Power line")

Selection and quality criteria
In order to support the selection of relevant studies that are aligned in the clarification of the proposed research questions, inclusion and exclusion criteria were defined. The inclusion of a study was determined by relevance, title analysis, summary, introduction, completion and when necessary by the complete reading of the study. The criteria adopted for exclusion and inclusion of the articles are presented in Tables 1 and 2. After the inclusion and exclusion criteria established, the next step was to define quality criteria. To analyze the quality of the primary studies, a list of five quality criteria for addressing the overall article has been defined and is presented in Table 3 based on those established by Kitchenham [4]. Publications that do not contain the exact key words or variations thereof will be discarded. EC2 Publications that do not use unmanned aircraft tools will be discarded. EC3 Publications that do not have the objective of monitoring transmission lines will be discarded. EC4 Publications that do not use the standalone mode will be discarded. EC5 Publications that do not have all the processes of the work or that are part of a work will be discarded. EC6 Publications that are only in planning or that have not been applied at least in the simulation environment will be discarded. Was the study described clearly and appropriately? QC2 Have the methods or techniques used in the primary studies been reported clearly? QC3 Was the virtual environment used in primary studies reported clearly? QC4 Was the study proposal evaluated/validated? QC5 Have the results been clearly stated?
In the criterion of note, to evaluate the articles, the scoring scale was used from 0 to 5, being that for each criterion attended completely was added 1 point, if partially 0.5 points and if it did not meet the criterion no score was added. For cost criteria, a qualitative assessment of the materials used was established, as not all-numerical information is accessible in the publications. The scale goes from very low to very high. UAVs manufactured from scratch and/or using only one piece of equipment will be downgraded to cost. Platelet UAVs and/or equipment add-ons will be enhanced in the cost rating. It is also evaluated in this criterion the cost of maintenance and resistance of the equipment from the information of the authors themselves. In the other criteria, a general analysis is performed, measuring from the information available in the publication itself. Section 4 presents the details of the arguments made by the respective authors.

CONDUCTION AND RESULTS
The implementation of the methodology presented in Figure 2 shows that the research returned a total of 223 articles (of these, 89 IEEE articles, 65 articles by Scopus, 5 by ACM and 64 Compendex articles), as the union of these articles had duplicates (85) and articles which were not aligned with the research questions (110), it was necessary to apply a screening process to obtain only relevant articles for research. The next step was to apply the inclusion and exclusion criteria resulting in 30 articles in total and finally a more careful selection was made by reading the articles completely, removing 4 articles with duplicate content, 5 false positive articles and 6 articles that the text complete was not available. Then 15 articles were obtained which are presented in the Table 4. After the final selection of the articles, the data extraction procedure was done from the complete reading of each article. The extracted data followed the specifications described below: -Publication data: title, author(s), publication date, publication vehicle and publication summary.
-General information: description of the objective of the article and materials used (to stipulate the cost). Each process is performed by two or three of the authors of this paper, this allows the review of decisions and eliminates any bias that may have occurred.
The results of this SRL were presented and analyzed in order to discuss the evidence from the 15 primary studies that may answer the research questions raised. As shown in Figure 3, the IEEE and Scopus database had a high representation with nine and eight included studies respectively, as for Compendex were five primary studies the ACM had less representation with no selected study. The systematic mapping, which is the summary of the analysis performed, is presented in the Table 4.  The most of articles studied proposed the use of UAV together with the camera for inspection. In general, in the article the UAV is manufactured by the team itself or with partnerships as in [9][10][11][12][13]21], which provides a reduced cost and better fits the researcher's needs as presented by the authors. Was an alternative in [8,13,14] adapt prefabricated or rack-mounted UAVs that generate a higher cost, mainly in maintenance. The authors of the articles [7,11,12,15] used, in addition to the basic material, LIDAR (light detection and variation) [22], a pulsed laser optical technology that measures light reflection to obtain distance from objects, and other laser types, considerably increase the cost of the solution, but increasing the accuracy of the laser line tracking. Its direction adjustment technique is to radiate the wire horizontally, obtaining line distance information at different locations on the wire. H1 and H2 in Figure 4 are the readings of the start and end distances. The angle (1) between the sides of the section and the central axis A approximates = 0 and H1 = H2, so that the UAV is parallel to the wire. In [8], the Kernelized Correlation Filter (KCF) algorithm [23] was used for locating and tracking between towers, this method uses rolling matrix properties to increase processing speed. This technique is known to be efficient, but costly to track lines. In [16] a combination of various aircraft was used and tracking was performed using a global positioning system (GPS) and inertial unit of measure (IMU) waypoint, however, the article does not guarantee accurate tracking as there is influences the effect of exposure of fields near power transmission towers [24].
In [10,14,[17][18][19] visual servoing type tracking [25] is used for UAV control, the method used is edge gradient detection. A smoothed image is filtered with the Sobel filter core [26] in the horizontal (Gx) and vertical (Gy) direction to obtain the first derivative (2) (3), then it traverses the pixels with a gradient vector that calculates the color intensity to identify the edge, as shown in Figure 5. Some articles use additional methods to improve the accuracy or performance of the algorithms. In [10,17] a robust Hough transformation method [27] is used to reduce noise and identify geometric shapes in images such as lines. In [10,14] also uses region of interest (ROI) separation algorithms to reduce the edge of the image on a given object, eliminating parts that should not be processed. This technique lets you explore the parameters of line continuity, similarity, and fixed spacing between power line structures.

239
It was proposed in [13,20] the use of magnetic sensors, the position and current parameters of the lines were reconstructed from the magnetic field data, with an approximation model, an inverse algorithm was proposed by the author and applied in the laboratory. In the geometrical model of the straight conductor of infinite length in the plane S of height H parallel to the horizontal plane is the power transmission line. The DC current flowing in the line is indicated as I and the intensity of the magnetic field is marked as A, and a vertical line of S is made through A and crosses the plane S at point B. A vertical line of L is made B to C. This model is represented in Figure 6 and the line equation in the plane (4). Figure 6. Transmission line model based on geometric relationship [20] Only in [12,16,21] is the energy efficiency of UAV addressed using other robot modes as shown in Figure 7. In [16] the use of different types of UAVs was discussed together and online for greater range but without losing the precision of the multirotor. In articles [12,21] the authors propose the use of suspended ground wire (OGW), a line support located at the highest part of the transmission tower, thus considerably increasing the range because it does not require much effort to maintain altitude. The problem here is that many towers do not have this type of installation accessible for use.

DISCUSSION AND IMPLEMENTATION
In this chapter, the results and the three main research questions will be discussed. In addition, in the end a general survey of the feasibility and limitations of the studies carried out are discussed.

Solution of questions
SQ1: Although the articles do not present current values used in the projects, it is possible to have notion of the cost of the implementation for the solution made. Maintenance cost decreases when using self-made materials. Articles that use a few equipment, mostly a camera and a sensor are inexpensive, while those using a variety of high-end equipment have a higher cost and the results, they get are similar to those. : All articles the technique is considered viable and has different methods, locations and equipment not being possible direct comparison. To stipulate efficiency of the technique used was considered the information of the author himself. Overall, the error rate, when presented, is around 10% thus demonstrating, thus considering the techniques employed as efficient. However, the energy question was little discussed, in general, the UAV had standard characteristics of autonomy.
SQ3: Due to the low error rate and the simulations or experiments carried out, it is assumed that UAVs are safe for application, but a specific approach to location and safety has not been presented in the articles. In general, the environments of use are in remote places and of little circulation of people, but in practice there are transmission lines that pass in the middle of great centers. Furthermore, no article has had a direct relationship to the safety of its implementations and equipment.

Simulation environment
Gazebo [28] was chosen as the simulation environment. The simulator can provide realistic feedback from the real world, with the accuracy of the dynamic environments a robot can encounter. The simulator also allows the use of angular and linear forces that can be applied to surfaces and joints to generate locomotion and interaction with the environment.
In this work, an object was used simulating a sequence of power transmission line towers of the freestanding delta model with circuit and robotic simulator UAV model Typhoon H480 with FPV, integrated with Gazebo through the PX4 simulator Figure 8. Initially, the UAV is on the ground parallel to the power transmission line, with its north to the final objective of the inspection, then to reach the desired height h in meters, a negative z-axis velocity is sent for a certain time t in seconds, as (1).
For line segment detection, the image needs to be properly handled. First, an image resizing to W (weight) 600 by H (height) 360 is performed, with the intention of speeding up the detection. Then the image is transformed to grayscale. The image is submitted to a Gaussian Blur filter with a Kernel value of 5, which allows highlighting the object of most interest [29].
The Canny edge detector is applied, turning the image into binary to identify the edges using neighboring pixels highlighted in white. Segmentation works with noise reduction through the Gaussian filter, then traverses the pixels with a gradient vector that calculates the color intensity to identify the edge as shown in the Figure 5.
The Hough Probabilistic algorithm is applied to this image treated with a specific region of interest, returning a list of identified segments. A filter is performed by eliminating segments with less than 25% H. Also, horizontal lines not in the range (6) are eliminated. After obtaining the valid segments, a check is made to find the largest and closest parallel segments by comparing by simple subtraction the two end points of the line on the x axis only, with the same thickness, and also grouping according to position. In negative response to these characteristics, it will be considered the first segment detected. The starting and ending points of the line are then returned along with the angle obtained. The line segments were identified in all cases in all environments (Figure 9), thus allowing the execution of the algorithm. However, the line length was identified at 95% in the simulator [30]. Figure 9. Image processing

Possible lacuna analysis
According to the information collected from the selected articles the low cost and efficient way is using a camera to use the algorithm edge detection that has little computational demand and a laser to adapt the algorithm supporting in the transmission line tracking.
The authors found that the solutions presented by the authors address the performance issue at a relatively low cost, but new embedded systems technologies for UAV appear at all times, a lacuna has been identified regarding the possibility of using deep learning with Reforcement Learning board [31]. For use on board the UAV, one can highlight the deep learning by Random Forest and Yolo, which allows for high accuracy with relatively low processing/training. These technologies can provide a generalization for different structures of towers in different environments.
For safety flight, it can be supplied with simple solutions such as parachutes, hook and protection of the propellers. It is also possible to develop redundancy systems for sensors and motors [32]. Vertical take-off and landing (VTOL) models, which are fixed wing multirotor hybrid electric UAVs, could be considered as an alternative for application by the authors for greater autonomy. Another possibility raised is the use of solar panels with [33], not necessarily for movement, but may be used for isolated components of the solution. These solutions were not presented in the articles.

CONCLUSION
Several technical means of tracking power transmission lines have been analyzed, proving to be a field that can be well explored, and with a lot of potential for development. Getting a low cost, widespread, efficient and secure solution, especially in a country with large extensions and limited financial resources in infrastructure, is a big challenge. The article in its direction obtained the resolution of the questions raised and identified possibilities not explored in the state of the art, with the focus on safety and efficiency.
It is intended by the authors the real implementation of autonomous tracking, using a university-made UAV, using solar panel, Lidar sensors and monocular camera. An adaptive algorithm for tower and line recognition will also be developed. There are reviews in the scientific literature about UAVs, however, being a systematic methodology of scientific review, no research has been found on this topic, which is very relevant in economic and social terms.